id	author	title	date	pages	extension	mime	words	sentence	flesch	summary	cache	txt
fcis-14697	Xu, Jingyu; Pan, Linying; Zeng, Qiang; Sun, Wenjian; Wan, Weixiang	Based on TPUGRAPHS Predicting Model Runtimes Using Graph Neural Networks	2023	4	.pdf	application/pdf	3206	192	50	Model results data Accuracy train test tiles 0.8785 0.8622 Layout:xla:random 0.6841 0.5285 Layout:xla:default 0.5631 0.5887 Layout:nlp:random 0.8197 0.8387 Layout:nlp:default 0.5036 0.4841 These results reinforce the potential of advanced deep learning techniques, especially GNN models, to improve the accuracy of model runtime predictions. Various recent methodologies leverage machine learning (ML) to acquire performance prediction models.	cache/fcis-14697.pdf	txt/fcis-14697.txt